Free electrons

7
In this chapter we investigate the optical properties associated with free
electrons. As the name suggests, these are systems in which the electrons
7.1 Plasma reflectivity 180 experience no restoring force from the medium when driven by the elec-
7.2 Free carrier tric field of a light wave. The two main solid-state systems that exhibit
conductivity 183
strong free electron effects are:
7.3 Metals 185
7.4 Doped • Metals. Metals contain large densities of free electrons that orig-
semiconductors 191 inate from the valence electrons of the metal atoms.
7.5 Plasmons 198
• Doped semiconductors. n-type semiconductors contain free elec-
7.6 Negative refraction 207 trons, while p-type materials contain free holes. The free carrier
Chapter summary 209 density is determined by the concentration of impurities used for
Further reading 210 the doping.
Exercises 211
We begin our discussion of the optical properties by using the Drude–
Lorentz model introduced in Section 2.1.3 of Chapter 2. This will enable
us to explain the main optical property of metals that we mentioned
in Section 1.4.3, namely that they reflect strongly in the visible spec-
tral region. We then apply our knowledge of interband transitions from
Chapter 3 to obtain a better understanding of the detailed form of the
reflectivity spectra of metals such as aluminium and copper. Next we ap-
ply the Drude–Lorentz model to doped semiconductors to explain why
doping causes infrared absorption. We then consider the collective oscil-
lations of the whole free carrier gas, which will naturally lead us to the
notion of plasmons, both bulk and surface. Finally, we conclude with a
brief discussion of negative refraction.

7.1 Plasma reflectivity

A neutral gas of charged particles is called a plasma. Metals and doped
semiconductors can be treated as plasmas because they contain equal
numbers of fixed positive ions and free electrons. The free electrons ex-
perience no restoring forces when they interact with electromagnetic
waves. This contrasts with bound electrons that have natural resonant
frequencies in the near-infrared, visible, or ultraviolet spectral regions
owing to the restoring forces of the medium.
In this section we derive a formula for the relative permittivity of
an electron plasma using the classical oscillator model discussed in Sec-
tion 2.2 of Chapter 2. As noted in Section 2.1.3, this approach com-
bines the Drude model of free electron conductivity with the Lorentz
 7.1 Plasma reflectivity 181

model of dipole oscillators, and is therefore known as the Drude–

Lorentz model.
We begin by considering the oscillations of a free electron induced by
the AC electric field E(t) of an electromagnetic wave. The equation of
motion for the displacement x of the electron is: We have assumed here that the light
is polarized along the x direction. The
d2 x dx model is not affected by this arbitrary
m0 + m0 γ = −eE(t) = −eE 0 e−iωt , (7.1) choice provided that the medium is
dt2 dt isotropic.
where ω is the angular frequency of the light, and E 0 is its amplitude. The
first term represents the acceleration of the electron, while the second is
the frictional damping force of the medium. The term on the right-hand
side is the driving force exerted by the light. Equation 7.1 is the same
as the equation of motion for a bound oscillator given in eqn 2.5, except
that there is no restoring force term because we are dealing with free
electrons.
By substituting x = x0 e−iωt into eqn 7.1, we obtain
eE
x= . (7.2)
m0 (ω 2 + iγω)
The polarization P of the gas is equal to −N ex, where N is the number
of electrons per unit volume. By recalling the definitions of the electric
displacement D and the relative permittivity r (cf. eqns A.2 and A.3),
we can write:
D = r 0 E
= 0 E + P
(7.3)
N e2 E
= 0 E − .
m0 (ω 2 + iγω)
Therefore:
N e2 1
r (ω) = 1 − . (7.4)
0 m0 (ω 2 + iγω)
This equation is identical to eqn 2.14 for the bound oscillator except that
the resonant frequency ω0 is zero and we have not yet considered the
effects of background polarizability. Equation 7.4 is frequently written
in the more concise form:
ωp2
r (ω) = 1 − , (7.5)
(ω 2 + iγω)
where
1/2
N e2
ωp = . (7.6)
0 m0 We shall see in Section 7.5 that ωp
corresponds to the natural resonant
ωp is known as the plasma frequency. frequency of the whole free carrier
Let us first consider a lightly damped system. In this case, we put gas. This contrasts with the resonant
γ = 0 in eqn 7.5 so that frequency of the individual electrons,
which is of course zero because they are
ωp2 free.
r (ω) = 1 − . (7.7)
ω2
182 Free electrons

1.0
0.8

Reflectivity
0.6
0.4
0.2
0.0
Fig. 7.1 Reflectivity of an undamped 0 1 2
free carrier gas as a function of fre- w / wp
quency.

The complex refractive index ñ of the medium is related to the complex

√
dielectric constant by ñ = r (cf. eqn 1.22). This means that ñ is
imaginary for ω < ωp and positive for ω > ωp , with a value of zero
The fact that the refractive index is precisely at ω = ωp . The reflectivity R can be calculated from eqn 1.29:
imaginary below ωp means that the ex-
tinction coefficient κ is large, and hence  
 ñ − 1 2
that the medium is highly attenuat- 
R=  . (7.8)
ing. This point will be explained further ñ + 1 
in Section 7.2. It is a general property
of systems with high extinction coeffi- By substituting the frequency dependence of ñ into this formula, we see
cients that they also have high reflec- that R is unity for ω ≤ ωp , and then decreases for ω > ωp , approaching
tivities.
zero at ω = ∞. This frequency dependence is plotted in Fig. 7.1.
The basic conclusion is that we expect the reflectivity of a gas of free
electrons to be 100% for frequencies up to ωp . This result is very well
confirmed by experimental data. In Sections 7.3 and 7.4 below we shall
see how the plasma reflectivity effect is observed in both metals and
doped semiconductors.

One of the best examples of plasma re- Example 7.1

flectivity effects is the reflection of ra-
dio waves from the upper atmosphere.
Aluminium is a trivalent metal with 6.0 × 1028 m−3 atoms per unit vol-
The atoms in the ionosphere are ion- ume. Account for the shiny appearance of aluminium.
ized by the ultraviolet light from the
Sun to produce a plasma of ions and Solution
free electrons. The plasma frequency
is in the MHz range, and so the low- Aluminium has three valence electrons per atom, and so the free elec-
frequency waves used for AM radio tron density N is 3 × (6.0 × 1028 ) = 1.8 × 1029 m−3 . We use this value
transmissions are reflected, but not the of N in eqn 7.6 to find that ωp = 2.4 × 1016 rad/s. The free electrons
higher-frequency waves used for FM ra- in aluminium will reflect all frequencies below ωp . Now ωp corresponds
dio or television. (See Exercise 7.2.)
to a wavelength of 2πc/ωp = 79 nm, which is in the ultraviolet spec-
tral region, and this means that all visible wavelengths are reflected.
Aluminium therefore has a shiny surface that can be used for making
mirrors.
 7.2 Free carrier conductivity 183

7.2 Free carrier conductivity

In deriving eqn 7.7, we neglected the damping of the free carrier oscil-
lations. We can recast the equation of motion in a way that makes the
physical significance of the damping term more apparent. To do this we
note that ẋ is the electron velocity v. Hence we can rewrite eqn 7.1 as:
dv
m0 + m0 γv = −eE . (7.9)
dt
Since m0 v is the momentum p, we see that:
dp p
= − − eE , (7.10)
dt τ
where we have replaced the damping rate γ by 1/τ , where τ is the damp-
ing time. This shows that the electron is being accelerated by the field,
but loses its momentum in time τ . In other words, τ is the momentum
scattering time.
In an AC field of the form E(t) = E 0 e−iωt , we look for solutions to
the equation of motion with x = x0 e−iωt . This implies that |v| = ẋ also
has a time variation of the form v = v 0 e−iωt . On substituting this into
eqn 7.9, we obtain:
−eτ 1
v(t) = E(t). (7.11)
m0 1 − iωτ
The current density j is related to the velocity and field through:

j = −N ev = σE , (7.12)

where σ is the electrical conductivity. On combining eqns 7.11 and 7.12,

we obtain the AC conductivity σ(ω):
σ0
σ(ω) = , (7.13)
1 − iωτ
where
N e2 τ
σ0 = . (7.14)
m0
σ0 is the conductivity measured with DC electric fields. We can thus de-
duce the momentum scattering time from the DC conductivity through
eqn 7.14. For a typical metal or doped semiconductor, this gives values
of τ in the range 10−14 to 10−13 s at room temperature.
By comparing eqns 7.4 and 7.13, we see that the AC conductivity and
the dielectric constant are related to each other through:
iσ(ω)
r (ω) = 1 + . (7.15)
0 ω
Thus optical measurements of r (ω) are equivalent to AC conductivity
measurements of σ(ω), and the free carrier reflectivity spectrum can be
discussed in terms of the conductivity rather than the dielectric constant.
184 Free electrons

At very low frequencies that satisfy ω  τ −1 , we can derive a useful

relationship between the conductivity of the free carrier gas and the
absorption coefficient for electromagnetic waves. This can be achieved by
first splitting r (ω) into its real and imaginary components in accordance
with eqn 1.21, i.e. r ≡ 1 + i2 . Equation 7.5 with γ = τ −1 gives:
ωp2 τ 2
1 = 1− (7.16)
1 + ω2 τ 2
ωp2 τ
2 = . (7.17)
ω(1 + ω 2 τ 2 )
We then work out n and κ, the real and imaginary parts of the complex
refractive index, by using eqns 1.25 and 1.26, and hence deduce the
absorption coefficient α from κ. Since ωτ  1 implies that 2
1 , we
can obtain solutions with n ≈ κ = (2 /2)1/2 . Using eqn 1.19 we therefore
obtain:  1/2
2ω(2 /2)1/2 2ωp2 τ ω
α= = . (7.18)
c c2
We can put this equation in a more accessible form by noting from
eqn 7.14 that ωp2 τ = σ0 /0 and from eqn A.28 that c2 = 1/0 µ0 . This
gives:
1/2
α = (2σ0 ωµ0 ) . (7.19)
Hence we see that the absorption coefficient is proportional to the square
root of the DC conductivity and the frequency.
Equation 7.19 implies that AC electric fields can only penetrate a short
distance into a conductor such as a metal. This well-known phenomenon
is called the skin effect. If the field strength varies as exp(−z/δ) with
the distance z from the surface, then the power falls off as exp(−2z/δ).
By comparison with the definition of α in eqn 1.4, we see that:

1/2
2 2
δ= = . (7.20)
α σ0 ωµ0
δ is known as the skin depth.
The fields that decay exponentially in the conductor are called evanes-
cent waves. We have seen in the previous section that we expect the
reflectivity of a metal to be very high for frequencies below ωp . From
what we have seen here, it is now apparent that this only applies if the
thickness l of the medium is much larger than the skin depth. If l is
comparable to, or smaller than, δ, the evanescent fields will not have
decayed fully by the back of the medium, and some of the energy will be
A treatment of the variation of R with transmitted. Conservation of energy then demands that the reflectivity
l may be found, for example, in Born R must drop accordingly. This implies that R depends on l when l  δ,
and Wolf (1999).
and ultimately drops to zero for a very thin medium.
At higher frequencies the relationship given in eqn 7.19 breaks down
because the approximation ωτ  1 is no longer valid. In this case we
can derive a different frequency dependence for the attenuation coeffi-
cient. This will be discussed when considering the absorption due to free
carriers in doped semiconductors in Section 7.4.1.
 7.3 Metals 185

Example 7.2
The DC electrical conductivity of copper is 6.5 × 107 Ω−1 m−1 at room
temperature. Calculate the skin depth at 50 Hz and 100 MHz.

Solution
The skin depth is given by eqn 7.20. At a frequency of 50 Hz we have
ω = 2π × 50 = 314 rad/s. On inserting this value of ω into eqn 7.20
and using σ0 = 6.5 × 107 Ω−1 m−1 , we obtain δ = 8.8 mm. At 100 MHz,
ω = 6.28 × 108 rad/s, and the skin depth δ is only 6.2 µm.

7.3 Metals
The free electron model of metals was proposed by P. Drude in 1900. The
model provides a basic explanation for why metals are good conductors
of heat and electricity, and is the starting point for more sophisticated
theories. As we shall see here, it is also successful in explaining why
metals tend to be good reflectors. On the other hand, band theory is
needed to explain why some metals (e.g. copper and gold) are coloured.

7.3.1 The Drude model

The Drude free electron model of metals considers the valence electrons
of the atoms to be free. When an electric field is applied, the free elec-
trons accelerate and then undergo collisions with the characteristic scat-
tering time τ introduced in eqn 7.10. The electrical conductivity is there-
fore limited by the scattering, and measurements of σ allow the value of
τ to be determined through eqn 7.14.
The free electron density N in the Drude model is equal to the den-
sity of metal atoms multiplied by their valency. Table 7.1 lists the Drude
free electron densities for a number of common metals. The values of N
are in the range 1028 –1029 m−3 . These very large free electron densities
explain why metals have high electrical and thermal conductivities. The
plasma frequencies calculated using eqn 7.6 are also tabulated in Ta-
ble 7.1, together with the wavelength λp that corresponds to ωp . It is
apparent that the very large values of N lead to plasma frequencies in
the ultraviolet spectral region.
In the visible spectral region where ω/2π ∼ 1015 Hz, we are usually
in a situation with ω
γ. This is because τ = γ −1 is typically of
order 10−14 s. Therefore the simplification of eqn 7.5 to eqn 7.7 is a
good approximation. With ωp in the ultraviolet, the visible photons have
frequencies below ωp and thus r is negative. As discussed in Section 7.1,
this means that the reflectivity is expected to be 100% up to ωp . This
explains the first and most obvious optical property of metals, namely
that they tend to be good reflectors at visible frequencies.
186 Free electrons

Table 7.1 Free electron density and plasma properties of

some metals. The figures are for room temperature unless
stated otherwise. The electron densities are based on data
taken from Wyckoff (1963). The plasma frequency ωp is
calculated from eqn 7.6, and λp is the wavelength corre-
sponding to this frequency.

Metal Valency N ωp /2π λp

(1028 m−3 ) (1015 Hz) (nm)

Li (77 K) 1 4.70 1.95 154

Na (5 K) 1 2.65 1.46 205
K (5 K) 1 1.40 1.06 282
Rb (5 K) 1 1.15 0.96 312
Cs (5 K) 1 0.91 0.86 350
Cu 1 8.47 2.61 115
Ag 1 5.86 2.17 138
Au 1 5.90 2.18 138
Be 2 24.7 4.46 67
Mg 2 8.61 2.63 114
Ca 2 4.61 1.93 156
Al 3 18.1 3.82 79

A striking implication of the free carrier model is that the dielec-

tric constant changes from being negative to positive as we go through
the plasma frequency. This means that the reflectivity ceases to be 100%
above ωp (see Fig. 7.1) and some of the light can be transmitted through
the metal. Thus we expect that all metals will eventually become trans-
mitting if we go far enough into the ultraviolet so that ω > ωp . This
phenomenon is known as the ultraviolet transparency of metals.
In order to observe the ultraviolet transmission threshold at the plasma
frequency, it is necessary that there should be no other absorption pro-
cesses occurring at ωp . This condition is best satisfied in the alkali met-
als. Table 7.2 lists the wavelengths of the ultraviolet transmission edges
Table 7.2 Ultraviolet transmission observed in the alkalis. The experimental wavelengths can be compared
threshold wavelength λUV for the alkali
metals. Data from Givens (1958). with those predicted from the calculated plasma frequency tabulated in
Table 7.1. The experimental results are in reasonable agreement with
Metal λUV (nm) the predictions, and show the correct trend on descending the periodic
table. The discrepancies can be explained to a large extent by replacing
Li 205 the free electron mass with the electron effective mass derived from the
Na 210
K 315
band structure of the metal. (See Exercise 7.4.)
Rb 360 Figure 7.2 shows the measured reflectivity of aluminium as a function
Cs 440 of photon energy from the infrared to the ultraviolet spectral region. As
noted in Example 7.1, the plasma frequency occurs in the ultraviolet
spectral region, and hence the reflectivity is expected to be high for all
visible frequencies. The data show that the reflectivity is over 80% for all
photon energies up to ∼ 15 eV, and then drops off to zero at higher ener-
gies. Thus aluminium shows the characteristic ultraviolet transparency
edge predicted by the Drude model. The relatively featureless reflectivity
at visible frequencies is exploited in commercial mirrors.
 7.3 Metals 187

1.0

0.8 hwp = 15.8 eV

Fig. 7.2 Experimental reflectivity of
Reflectivity

experimental data aluminium as a function of photon en-

0.6 ergy. The experimental data are com-
pared to predictions of the free electron
0.4 g=0 model with ~ωp = 15.8 eV. The dot-
ted curve is calculated with no damp-
0.2 t = 8.0 fs ing. The dashed line is calculated with
τ = 8.0 × 10−15 s, which is the value
0.0 deduced from DC conductivity. Exper-
0 5 10 15 20 imental data from Ehrenreich et al.
Energy (eV) (1963),
c American Physical Society,
reprinted with permission.

The plasma frequency for aluminium listed in Table 7.1 corresponds

to a photon energy of 15.8 eV. The dotted line in Fig. 7.2 gives the
reflectivity predicted from eqn 7.7 with
ωp = 15.8 eV. On comparing We shall see in Section 7.5.1 that the
the experimental and theoretical results, we see that the model accounts plasma frequency can be determined
directly by using electron energy-loss
for the general shape of the spectrum, but there are some important spectroscopy.
details that are not explained.
An improved attempt to model the experimental data can be made
by including the damping term in the dielectric constant. Example 7.3
explains how this is done. The reflectivity calculated from eqn 7.5 for
the value of τ deduced from the DC conductivity, namely 8.0 × 10−15 s,
is plotted as the dashed line in Fig. 7.2. The main difference between
the two calculated curves is that the damping causes the reflectivity
to be less than unity below ωp , and the ultraviolet transmission edge is
slightly broadened. However, this is only a relatively small effect because
ωp
τ −1 .
The inclusion of damping makes a small improvement in the fit to the
data, but there are two important features that are still not explained.
Firstly, the reflectivity is significantly lower than predicted, and secondly,
there is a dip around 1.5 eV, where we would have expected a featureless
curve. Both of these points can be explained by considering the interband
absorption rates. These are discussed in the next section.

Example 7.3
The conductivity of aluminium at room temperature is 4.1×107 Ω−1 m−1 .
Calculate the reflectivity at 500 nm according to the Drude–Lorentz
model.

Solution
We first work out the damping time τ from the conductivity using
eqn 7.14. Taking the value of N = 1.81 × 1029 m−3 from Table 7.1,
we find:
m0 σ 0
τ= = 8.0 × 10−15 s .
N e2
188 Free electrons

Table 7.1 also gives us the value of the plasma frequency, namely ωp =
2.4 × 1016 rad/s. The wavelength of 500 nm corresponds to an angular
frequency ω = 2πc/λ = 3.8 × 1015 rad/s. We use these frequencies in
eqns 7.16 and 7.17 to calculate the real and imaginary parts of the
complex dielectric constant:
ωp2 τ 2
1 = 1 − = −39 ,
1 + ω2 τ 2
and
ωp2 τ
2 = = 1.3 .
ω(1 + ω 2 τ 2 )
We then work out the real and imaginary parts of the complex refractive
index using eqns 1.25 and 1.26. This gives:
1  1/2
n = √ − 39 + [(−39)2 + (1.3)2 ]1/2 = 0.10 ,
2
and
1  1/2
κ = √ + 39 + [(−39)2 + (1.3)2 ]1/2 = 6.2 .
2
We finally obtain the reflectivity from eqn 1.29:
(n − 1)2 + κ2 (−0.9)2 + (6.2)2
R= = = 99% .
(n + 1)2 + κ2 (1.1)2 + (6.2)2
This shows that the inclusion of the damping reduces the reflectivity by
only 1% in this case.

7.3.2 Interband transitions in metals

The absorption of light by direct interband transitions was discussed in
detail in Chapter 3. Direct transitions involve the promotion of electrons
to a higher band by absorption of a photon with the correct energy. The
electron does not change its k vector significantly because of the very
small momentum of the photon. Thus the transitions appear as vertical
arrows on the E–k band diagram of the solid.
Interband absorption is important in metals because the electromag-
netic waves penetrate a short distance into the surface, and if there is
a significant probability for interband absorption, the reflectivity will
be reduced. We consider below the reflectivity spectra of aluminium and
copper in order to illustrate the effects of interband absorption, and then
make some general comments about other metals such as silver and gold.

Aluminium
The band diagram of aluminium is shown in Fig. 7.3. Aluminium has an
electronic configuration of [Ne]3s2 3p1 with three valence electrons. The
crystal structure is face-centred cubic, which has a body-centred cubic
 7.3 Metals 189

16
Al
Energy (eV)

12 EF

8
Fig. 7.3 Band diagram of aluminium.
4 The transitions at the W and K points
that are responsible for the reflectivity
dip at 1.5 eV are labelled. After Segall
0
G X W L G K X (1961),
c American Physical Society,
reprinted with permission.

(bcc) reciprocal lattice, as shown in Fig. D.5 of Appendix D. The first

Brillouin zone is completely full, and the valence electrons spread into
the second, third and slightly into the fourth zones. The band structure
appears quite complex due to the irregular shape of the bcc Brillouin
zone. However, the bands are actually very close to the free electron
model, with significant departures only in the vicinity of the Brillouin
zone boundaries. The bands are filled up to the Fermi energy EF , which
is marked on the diagram. Direct transitions can take place from any
of the states below the Fermi level to unoccupied bands directly above
them on the E–k diagram.
Fermi’s golden rule given in eqn 3.2 tells us that the absorption rate
is proportional to the density of states for the transition. The dip in the
reflectivity at 1.5 eV which is apparent in Fig. 7.2 is a consequence of
We came across a similar example of
the ‘parallel band’ effect. This occurs when there is a band above the parallel bands when we discussed the
Fermi level that is approximately parallel to another band below EF . In absorption rate at the critical points
this case, the interband transitions from a large number of occupied k in the band structure of silicon in Sec-
states below the Fermi level will all occur at the same energy. Hence the tion 3.5.
density of states at the energy difference between the two parallel bands
will be very high, which will result in a particularly strong absorption
at this photon energy.
Inspection of the band diagram of aluminium shows that the parallel
The positions of the W and K points
band effect occurs at both the W and K points of the Brillouin zone. on the fcc Brillouin zone boundary are
These transitions have been identified on Fig. 7.3. The energy separa- shown in Fig. D.5.
tion of the parallel bands is approximately 1.5 eV in both cases. The
enhanced transition rate at this photon energy thus explains the reflec-
tivity dip observed at 1.5 eV in the experimental data. Moreover, we can
see from the band diagram that there will be further transitions between
bands below the Fermi level to unoccupied bands above EF at a whole
range of photon energies greater than 1.5 eV. The density of states for
these transitions will be lower than at 1.5 eV because the bands are not
parallel. However, the absorption rate is still significant, and accounts
for the reduction of the reflectivity to a value below that predicted by
the Drude model in the visible and ultraviolet spectral regions.
190 Free electrons

Copper
Copper has an electronic configuration of [Ar]3d10 4s1 . The outer 4s
bands approximate reasonably well to free electron states with a dis-
persion given by E =
2 k 2 /2m0 . They therefore form a broad band
covering a wide range of energies. The 3d bands, on the other hand,
are more tightly bound and are relatively dispersionless, occupying only
a narrow range of energies. The density of states for the two bands is
illustrated schematically in Fig. 7.4. The narrow 3d bands can hold ten
electrons, and therefore their density of states is sharply peaked. The
4s band, which can hold two electrons, is much broader, with a smaller
maximum. The 11 valence electrons of copper fill up the 3d band, and
half fill the 4s band. The Fermi energy therefore lies within the 4s band
In atomic physics, transitions between above the 3d band. Interband transitions are possible from the filled
d and s states are forbidden for electric- 3d bands to unoccupied states in the 4s band above EF , as illustrated
dipole processes. (See Table B.1 in Ap- in Fig. 7.4. This implies that there will be a well-defined threshold for
pendix B.) The matrix element for
the 3d→4s transitions is therefore rel- interband transitions from the 3d bands to the 4s band.
atively small, but this is compensated Figure 7.5 shows the actual band structure and density of states of
by the very high density of states in copper. The general features indicated in Fig. 7.4 are clearly shown in
the solid, which results in strong ab- the calculated curves. The 4s band is the parabola starting at the Γ point
sorption.
at −9 eV, while the 3d bands are the five curves bunched together in the
energy range −5 → −2 eV. The 4s band crosses the 3d bands and then
re-emerges as the single band with energy > −2 eV. It is apparent that
the 3d electrons lie in relatively narrow bands with very high densities
E of states, while the 4s band is much broader with a lower density of
states. The Fermi energy lies in the middle of the 4s band above the
4s band 3d band. Interband transitions are possible from the 3d bands below
EF to unoccupied levels in the 4s band above EF . The lowest energy
optical transitions are marked on the band diagram in Fig. 7.5. The transition
transitions energy is 2.2 eV which corresponds to a wavelength of 560 nm.
EF Figure 7.6 shows the measured reflectivity of copper from the infrared
to the ultraviolet spectral region. Based on the plasma frequency given
in Table 7.1, we would expect near 100% reflectivity for photon energies
below 10.8 eV, which corresponds to an ultraviolet wavelength of 115 nm.
3d band
However, the experimental reflectivity falls off sharply above 2 eV owing
to the interband absorption edge discussed above. This explains why
Density of states
copper has a reddish colour.

Fig. 7.4 Schematic density of states Silver and gold

for the 3d and 4s bands of a transition
metal such as copper. The arguments used for copper can be applied to the other noble metals.
The important parameter is the energy gap between the d bands and
the Fermi energy, as shown in Fig. 7.4. In gold the interband absorption
threshold occurs at a slightly higher energy than copper, which explains
why it has a yellowish colour. In silver, on the other hand, the interband
absorption edge is around 4 eV. This frequency is in the ultraviolet, and
so the reflectivity remains high throughout the whole visible spectrum.
(See Fig. 1.5.) This explains why silver does not have any particular
colour, and also why it is so widely used for making mirrors. Gold is also
 7.4 Doped semiconductors 191

Integrated density of states

0 2 4 6 8 10
2 4s copper 4s 2
0 EF 0
Energy (eV)

Energy (eV)
-2 -2
3d
-4 bands -4
-6 -6
copper
-8 3d and 4s bands -8
4s band
W L G X W K 0 2 4 6
Density of states (states eV-1)
Wave vector k

Fig. 7.5 Calculated band structure of copper. The transitions from the 3d bands responsible for the interband transitions
around 2 eV are identified. The right-hand side of the figure shows the density of states calculated from the band structure. The
strongly peaked features between about −2 eV and −5 eV are due to the 3d bands. The dotted line is the integrated density of
states. The Fermi level (defined here as E = 0) corresponds to the energy at which the integrated density of states is equal to
11. After Moruzzi et al. (1978).

used for mirrors, but only at infrared wavelengths.

7.4 Doped semiconductors

The controlled doping of semiconductors with impurities is an essential In silicon and germanium, which come
part of solid-state technology. The general principles are discussed in from group IV of the periodic table,
n-type doping is achieved by adding
Section D.1 of Appendix D. The introduction of donor impurities pro- atoms from group V, while p-type dop-
vides an excess of electrons, while acceptor impurities lead to a deficit of ing is achieved by adding atoms from
electrons, which is equivalent to an excess of holes. Doping that produces group III. In compound semiconductors
excess electrons is called n-type, while doping that produces excess holes such as the III–Vs, the way the doping
works is more complicated. If the impu-
is called p-type. rity sits on the group III atom site, then
Experimental measurements on doped semiconductors show that the a group II element gives p-type doping,
presence of impurities gives rise to new absorption mechanisms and also and a group IV element gives n-type.
On the other hand, if the impurity sits
to a free carrier plasma reflectivity edge. Our aim here is to explain these
on the group V atom site, then a group
effects by applying a suitably modified version of the free carrier model IV impurity gives p-type doping, while
and by considering the quantized levels created by the impurity atoms. In a group VI element gives n-type. Group
the two subsections that follow, we first consider the free carrier effects, IV doping of a III–V semiconductor can
therefore give either n- or p-type dop-
and then move on to discuss the absorption associated with the impurity ing, depending on how the impurities
levels. fit into the crystal.

7.4.1 Free carrier reflectivity and absorption

The free electron model developed in Sections 7.1 and 7.2 can be ap-
plied to doped semiconductors if we make two appropriate modifica-
tions. Firstly, we must account for the fact that the electrons and holes
192 Free electrons

1.0

0.8 copper

Reflectivity
0.6

0.4
hwp

visible
0.2
Fig. 7.6 Reflectivity of copper from
the infrared to the ultraviolet spectral
region. The reflectivity drops sharply 0.0
above 2 eV due to interband transi-
0 2 4 6 8 10 12 14
tions. Data from Lide (1996). Energy (eV)

are moving in the conduction or valence band of a semiconductor. This

is easily achieved by assuming that the carriers behave as particles with
an effective mass m∗ rather than the free electron mass m0 . Secondly, we
must remember that the semiconductor has a high relative permittivity
at the frequencies of interest even before the dopants are added.
The two modifications mentioned above can be handled if we rewrite
eqn 7.3 in the following form:
D = r 0 E
= 0 E + Pother + Pfree carrier
(7.21)
N e2 E
= opt 0 E − ∗ 2 .
m (ω + iγω)
The term Pother accounts for the polarizability of the bound electrons
before the dopants are added, while the effective mass m∗ accounts for
the band structure of the semiconductor. The carrier density N that
appears in this equation is the density of free electrons or holes generated
by the doping process. Note that the sign of the charge cancels, and so
As explained in Section 2.2.2 of Chap- the only difference between electrons and holes in this treatment is in
ter 2, solids have a number of resonant
frequencies, each of which can be mod-
the effective mass that is used.
elled by dipole oscillators. There are The free carrier effects due to doping are most noticeable in the spec-
resonant frequencies in the infrared due tral region 5–30 µm, where we would normally expect the semiconductor
to the phonons, and others in the near- to be completely transparent. Hence the value of opt that we use in
infrared, visible, or ultraviolet due to
the bound electrons. The phonon ab-
eqn 7.21 is the one measured in the transparent spectral region below
sorption bands are discussed in detail the interband absorption edge. This value is known from the refractive
in Chapter 10, and occur in the range index of the undoped semiconductor: opt = n2 . (See eqn 1.27 with κ = 0
30–100 m for a typical III–V semicon- below the band edge.)
ductor.
Equation 7.21 tells us that the frequency dependence of the dielectric
constant is given by:
N e2 1
r (ω) = opt − . (7.22)
m∗ 0 (ω 2 + iγω)
This can be rewritten as:
 
ωp2
r (ω) = opt 1− 2 , (7.23)
(ω + iγω)
 7.4 Doped semiconductors 193

100
n-type InSb
80 4.0 ´ 1024
Reflectivity (%)

2.8 ´ 1024
1.2 ´ 1024
60 6.2 ´ 1023
3.5 ´ 1023
40

20
Fig. 7.7 Infrared reflectivity spectra of
n-type InSb at room temperature for
0 different values of the free electron den-
10 20 30 sity. After Spitzer & Fan (1957),
c
Wavelength (mm) American Physical Society, reprinted
with permission.

where the plasma frequency ωp is now given by:

N e2
ωp2 = . (7.24)
opt 0 m∗
We have written the dielectric constant in this way to make the link to
the Drude model apparent. The difference between the plasma frequency
for the semiconductor given in eqn 7.24 and that given in eqn 7.6 is that
we have replaced m0 by m∗ , and we have included opt to account for
the background polarizability of the undoped semiconductor.
If we assume that the system is lightly damped, then we can ignore the
damping term in eqn 7.23. This then implies that r is negative below ωp
and positive at higher frequencies. We thus expect to observe a plasma
reflectivity edge at ωp just as we did in metals. Since the carrier density
is much smaller than in metals, the plasma edge occurs at frequencies
in the infrared spectral range. This prediction is very well borne out by
infrared reflectivity data.
Figure 7.7 shows the measured reflectivity of n-type InSb as a func-
tion of the electron density. The fundamental absorption edge at the
band gap of InSb occurs at 6 µm, while the phonon absorption band lies
around 50 µm. Thus we would expect pure InSb to be transparent in the
wavelength range shown and have a featureless reflectivity spectrum.
Instead, the data show a well defined reflectivity edge, which shifts to
shorter wavelengths as the electron density increases, in accordance with
eqn 7.24.
The data shown in Fig. 7.7 demonstrate the phenomenon of the plasma
reflectivity edge more clearly than many of the equivalent results ob-
tained for metals. This is because it is not possible to vary the electron
density in metals. Moreover, in metals the plasma frequencies are much
higher, and the reflectivity edge is frequently obscured by interband
transitions.
One very striking feature of the data is the zero in the reflectivity
at wavelengths just below the plasma edge. This occurs at a frequency
194 Free electrons

given by (see Exercise 7.8):

opt
ω2 = ω2 . (7.25)
opt − 1 p

By fitting this formula to the data, the effective mass of InSb can be
determined. (See Exercise 7.9.)
At frequencies above ωp , the presence of free carriers leads to the ab-
sorption of light. This effect is called free carrier absorption, and
can be observed in the infrared spectral region below the fundamental
absorption edge at the band gap, where the semiconductor would nor-
mally be transparent. To see how this effect arises, we split the dielectric
constant given in eqn 7.23 into its real and imaginary parts. This gives:
 
ωp2 τ 2
1 = opt 1 − , (7.26)
1 + ω2τ 2
opt ωp2 τ
2 = , (7.27)
ω(1 + ω 2 τ 2 )

where we have made the usual substitution of τ −1 for γ. In a typical

semiconductor, with τ ∼ 10−13 s at room temperature, it is safe to make
the approximation ωτ
1 at frequencies in the near-infrared. Further-
more, the free carrier term in r will be small. Therefore we can assume
1 ≈ opt , and that 2  1 . In these conditions we find solutions to
The skin effect considered in Section 7.2 √
eqns 1.25 and 1.26 with n = opt and κ = 2 /2n. This allows us to
may also be considered as a type of free
carrier absorption. In the skin effect,
deduce the absorption coefficient using eqn 1.19. The result is:
however, we are considering the absorp-
tion at low frequencies below ωp where opt ωp2 N e2 1
the material is highly reflective. We are
αfree carrier = 2
= ∗ . (7.28)
ncω τ m 0 ncτ ω 2
now considering absorption above ωp
where the material should be transpar- This shows that the free carrier absorption is proportional to the carrier
ent.
density and should vary with frequency as ω−2 .
Experimental data on a number of n-doped samples lead to the conclu-
sion that αfree carrier ∝ ω −β , where β is in the range 2–3. The departure
of β from the predicted value of 2 is caused by the failure of our as-
The approximation that τ is indepen- sumption that τ is independent of ω. To see why this is important, we
dent of ω effectively says that the re-
laxation time of the electrons does not illustrate the physical processes that are occurring during free carrier ab-
depend on their initial energy. This sorption in Fig. 7.8. The figure shows the conduction band of an n-type
is equivalent to the energy indepen- semiconductor, which is filled up to the Fermi level determined by the
dent relaxation time approximation of
free carrier density. Absorption of a photon excites an electron from an
the Boltzmann equation used in elec-
tron transport theory. It is well known occupied state below the Fermi level to an unoccupied level above EF .
that this approximation is only valid The photon only has a very small momentum compared to the electron,
in a limited range of conditions. See and therefore cannot change the electron’s momentum significantly. It
Ashcroft and Mermin (1976) for further
details.
is obvious from Fig. 7.8 that a scattering event must occur to conserve
momentum in the process. Hence the absorption must be proportional
to the scattering rate 1/τ , in accordance with the prediction of eqn 7.28.
The mechanisms that can contribute to the momentum-conserving
process in free carrier absorption include phonon scattering and scat-
tering from the ionized impurities left behind by the release of the free
 7.4 Doped semiconductors 195

E
hk 2
E= scattering
2me*

hw
EF

k Fig. 7.8 A free carrier transition in a

doped semiconductor.

E
Fig. 7.9 Intervalence band absorption
 ‚ ƒ in a p-type semiconductor. EF is the
Fermi energy determined by the dop-
k
D EF ing density. The labelled arrows indi-
cate: (1) transitions from the light-hole
hh band (lh) band to the heavy-hole (hh) band;
lh band (2) transitions from the split-off (SO)
SO band
band to the lh band; and (3) transitions
from the SO band to the hh band.

electrons from their dopants. It is a sweeping oversimplification to char-

acterize all the possible scattering processes with a single frequency-
independent scattering time τ deduced from the DC conductivity. Thus
it is hardly surprising that the experimental data do not exactly show
an ω −2 dependence.
The free carrier reflectivity and absorption of p-type semiconductors
can be modelled by a similar treatment to the one developed here for
n-type samples. The only change that has to be made is in the effective
mass that is used in the calculation. Thus we would expect that all
the main results will hold, provided we take account of the fact that the
scattering time for holes is not necessarily the same as that for electrons.
However, p-type samples also show another effect, which is discussed
next.
Figure 7.9 shows the valence band of a p-type III–V semiconductor.
This is a larger scale version of the band structure diagram given pre-
viously in Fig. 3.5, except that now there are unfilled states near k = 0
owing to the p-type doping. Optical transitions can take place in which
an electron is promoted from an occupied state below EF in the light-hole
(lh) band to an empty one in the heavy-hole (hh) band above EF . This is
called intervalence band absorption. Other intervalence band tran- Note that intervalence band transitions
sitions are possible in which an electron is promoted from the split-off are forbidden at k = 0 since all the hole
bands are derived from p-like atomic
(SO) band to either the lh or hh band. The range of energies over which states. The atomic character of the
these transitions occur can be calculated from the effective masses, the bands is less well defined for finite k,
doping density and the spin–orbit energy ∆ (see Exercise 7.12). The ab- and this makes the transitions possible
sorption occurs in the infrared, and measurements of the spectrum can away from the centre of the Brillouin
zone.
give values for ∆ and the ratio of the hole effective masses. The absorp-
tion can be a strong process because no scattering events are required
196 Free electrons

conduction band conduction band

n n
3 3
2 donor 2
levels
1 1

Fig. 7.10 Impurity absorption mech-

anisms in an n-type semiconductor:
(a) transitions between donor levels;
(b) transitions from the valence band to
empty donor levels. The donor level en- valence band valence band
ergy spacings have been exaggerated in
this diagram to make the mechanisms (a) (b)
clearer.

to conserve momentum.

7.4.2 Impurity absorption

The n-type doping of a semiconductor with donor atoms introduces a
series of hydrogenic levels just below the conduction band. These quan-
tized states are called donor levels, and are illustrated schematically
in Fig. 7.10. The impurity levels give rise to two new absorption mech-
anisms, in addition to the free carrier effects discussed in the previous
section. If the donor states are occupied, it will be possible to absorb pho-
tons by exciting electrons between the levels as illustrated in Fig. 7.10(a).
On the other hand, if the states are empty, then it will be possible to
absorb light by exciting electrons from the valence band to the donor
states as illustrated in Fig. 7.10(b).
We consider first the transitions between the donor levels illustrated
in Fig. 7.10(a). For such a process to occur, the donor levels must be
occupied. This will be the case at low temperatures, when there is in-
sufficient thermal energy to promote the electrons from the donor levels
into the conduction band.
The frequencies of the donor-level transitions can be calculated if the
energies of the impurity states are known. In the simplest model, we
assume that the electron is released into the crystal, and is then attracted
back towards the positively charged impurity atom. The electron and
Equation 7.29 is very similar to eqn 4.1 the ionized impurity then form a hydrogenic system bound together by
for the exciton binding energy except their mutual Coulomb attraction. As a first approximation, we can use
that the electron effective mass ap- the Bohr formula, provided we use the effective mass m∗e instead of the
pears instead of the reduced electron-
free electron mass m0 , and also include the dielectric constant r for the
hole mass. This is because we are now
considering the attraction of an elec- semiconductor. Hence the energy of the donor levels EnD will be given
tron to a heavy ion which is bound in by:
the lattice, instead of that between a m∗ 1 RH
free electron and a free hole. EnD = − e 2 2 , (7.29)
m0 r n
 7.4 Doped semiconductors 197

1.0
Absorption (103 m-1) 2p±
2p0
3p±
Si : P
T = 4.2 K 4p±
3p0 5p± Fig. 7.11 Infrared absorption spec-
4p0 trum of n-type silicon doped with phos-
phorus at a density of 1.2 × 1020 m−3 .
0 The temperature was 4.2 K. After Ja-
32 34 36 38 40 42 44 gannath et al. (1981),
c American
Photon energy (meV) Physical Society, reprinted with per-
mission.

where RH is the hydrogen Rydberg energy (13.6 eV) and n is an integer.

At low temperatures we can assume that all the electrons from the
donors will be in the n = 1 ground state impurity level. Optical transi-
tions can then take place in which the electrons are promoted to higher
donor levels or into the conduction band by absorption of a photon.
Figure 7.10(a) illustrates two possible transitions of this type, in which
the electron is promoted to either the n = 2 or the n = 3 donor level.
These transitions give rise to absorption lines analogous to the hydrogen
Lyman series with frequencies given by:


m∗ RH 1
hν = e 2 1− 2 , (7.30)
m0 r n
where n is the quantum number of the final impurity level. If we insert
typical values into eqn 7.30 we find that the photon energies are in the
range 0.01–0.1 eV. This means that the transitions occur in the infrared
spectral region.
Figure 7.11 shows the absorption spectrum of n-type silicon at liquid The absolute value of the absorption
helium temperatures. The sample was doped with phosphorus at a den- coefficient for the impurity transitions
sity of 1.2 × 1020 m−3 . The absorption lines correspond to transitions in Fig. 7.11 is around 103 m−1 . This
is much smaller than for interband
exciting electrons from the n = 1 shell to higher shells. In the language transitions which typically have val-
of atomic physics, these are 1s → np transitions. These transitions con- ues in the range 106 –108 m−1 . How-
verge at high n to the donor ionization energy of phosphorus in silicon, ever, if we were to assume that the
which is 45 meV. absorption strength is simply propor-
tional to the number of atoms that con-
The spectrum shown in Fig. 7.11 is actually more complicated than tribute, we would expect the impurity
eqn 7.30 would suggest. It consists of two series of transitions, which absorption to be weaker than the in-
are labelled as either np0 or np± . The np0 series obey eqn 7.30 very terband absorption by about a factor
well, but the np± transitions have a different frequency dependence. of ∼ 109 . The measured ratio is much
larger because the impurity lines are
This complexity is caused by the anisotropy of the effective mass of very sharp, whereas the interband tran-
silicon. The frequency dependence of the two series can be modelled by sitions spread out into bands.
assigning different effective Rydberg energies for the ‘0’ and ‘±’ states.
(See Exercise 7.13.)
We now consider the absorption mechanism shown in Fig. 7.10(b).
These transitions can be observed at temperatures when the donor levels
are partly unoccupied owing to the thermal excitation of the electrons
into the conduction band. Absorption processes can then occur in which
198 Free electrons

electrons are excited from the top of the valence band to the empty
donor levels.
The valence band → donor level transitions occur at photon energies
In many direct-gap semiconductors it is
just below the band gap Eg , with a threshold given by Eg −E1D . However,
found experimentally that the absorp- the transitions tend to be broadened into a continuum both by thermal
tion decreases exponentially below the effects and by the fact that transitions can take place from a whole
band gap. This is called the Urbach tail range of states within the valence band. Hence the impurity transitions
on account of Urbach’s rule, which
states that the frequency dependence of cause a smearing of the absorption edge compared with the abrupt edge
the absorption for ~ω < Eg is given by: found at the band gap of pure semiconductors. The absorption strength
 σ(~ω − E ) will always be weak compared to the interband and excitonic transitions
g
α(~ω) ∝ exp , due to the relatively small number of impurity atoms compared to the
kB T
where σ is a phenomenological fitting
density of states within the conduction band. On the other hand, the
parameter. transitions occur in the spectral region just below the band gap where
we would normally expect no absorption at all. Hence these transitions
do have an effect on the fundamental absorption edge, and make precise
determinations of Eg from the absorption spectra at room temperature
more difficult.
We have restricted our attention here to n-type semiconductors for
the sake of simplicity. The same effects can of course occur in p-type
materials.

7.5 Plasmons
Equation 7.7 tells us that the relative permittivity of a lightly damped
gas of free electrons is expected to be zero at ωp . This suggests that
something interesting might happen at this frequency. This is indeed
the case, as we discuss here.

7.5.1 Bulk plasmons

A plasma consists of a gas of charged particles in dynamic equilibrium.
The particles are in constant motion, and this can create local charge
fluctuations. If a fluctuation were to create a small region with an excess
charge, the charges in that volume would be repelled away by the sur-
rounding charges. The velocity acquired in this process could cause the
excess charges to overshoot their original position, in which case they
would then be pushed back in the opposite direction. This process can
lead to oscillatory motion called plasma oscillations. These plasma os-
cillations are well known in gas discharge tubes, and they can also occur
in the free electron plasmas found in metals and doped semiconductors,
which is our interest here.
The frequency of the plasma oscillations can be calculated as follows.
Consider a region of a conducting medium of volume V enclosed by a
surface S. Conservation of charge requires that the net flow of current
into or out of a region must be balanced by the change of the total charge
 7.5 Plasmons 199

inside the surface. This continuity condition can be written:

 
∂
j · dS = − ρ dV , (7.31)
S ∂t V
where j is the current density, dS is a surface element, ρ is the local
charge density, and dV is a volume element. On applying the divergence The divergence theorem of mathemat-
theorem, we find that: ics requires that:
I Z
 
∂ρ j
dS = H
j dV ,
∇ · j dV = − dV , (7.32) S V
V V ∂t where the volume integral is over the
region enclosed by the surface S.
and hence (since the volume integrated over is arbitrary):
∂ρ
∇·j =− . (7.33)
∂t
Equation 7.33 is called the charge continuity equation.
We now consider the case in which we have a collective motion of
the free electrons relative to the fixed lattice of positive ions in a metal
or doped semiconductor. The overall charge density is zero, but the
movement of the electrons can create local currents. Since the positive
charges on the ions are stationary, they do not generate a current, and we
can apply the charge continuity equation to the electron current alone,
giving
∂ρe
∇·j =− , (7.34)
∂t
where ρe is the electron charge density. Then, on substituting for ρe from
Gauss’s law (i.e. ∇ · E = ρe /0 ), we find:


∂E
∇· j + 0 = 0. (7.35)
∂t
Now a vector that has zero divergence can always be written as the curl
of another vector, and from Maxwell’s fourth equation (eqn A.13), we
realize that this vector must be B/µ0 , giving:
∂E 1
j + 0 = ∇×B. (7.36)
∂t µ0
On taking the time derivative and substituting from Maxwell’s third
equation (eqn A.12), we find:

∂j ∂2E 1 ∂B
+ 0 2 = ∇×
∂t ∂t µ0 ∂t
(7.37)
1
=− ∇× (∇ × E) .
µ0
The electrons will move in response to the local electric field according
to their equation of motion:

mv̇ = −eE . (7.38)

200 Free electrons

Now the current density is given by j = −N ev, which implies that

∂j N e2
= E. (7.39)
∂t m
On substituting into eqn 7.37 and re-arranging, we find:
∂2E
+ ωp2 E = −c2 ∇× (∇ × E) , (7.40)
∂t2
where we have substituted for ωp from eqn 7.6 and used c2 = 1/µ0 0 (cf.
eqn A.28).
At this stage it is helpful to split the electric field into transverse and
longitudinal components:

E = Et + El , (7.41)

where ∇ · E t = 0 and ∇ × E l = 0. On substituting into eqn 7.40 and

using the vector identity of eqn A.24, this gives:

2 
∂ 2E t ∂ El
+ ωp E t − c ∇ E t = −
2 2 2
+ ωp E l .
2
(7.42)
∂t2 ∂t2

It becomes apparent that both sides of This implies that we have two independent equations of motion for the
eqn 7.42 must be equal to zero by tak- transverse and longitudinal components:
ing the divergence and curl of the equa-
tion. ∂ 2E t
+ ωp2 E t − c2 ∇2 E t = 0, (7.43)
∂t2
∂2E l
+ ωp2 E l = 0. (7.44)
∂t2
We look for wave-like solutions in which the field varies with time and
position as exp i(k · r − ωt). For the transverse solutions we find:

c2 k 2 = ω 2 − ωp2 . (7.45)

This describes the dispersion of conventional transverse electromagnetic

waves in the plasma. The dispersion is plotted in Fig. 7.16(a) below.
There are no travelling solutions with ω < ωp because the waves are
reflected by the plasma.
In the case of the longitudinal modes, we simply have:

ω = ωp . (7.46)

This shows that the medium can support longitudinal modes at the
In a metal, the frequency of the longi- plasma frequency, and that the modes are dispersionless: i.e. ω is in-
tudinal modes does in fact vary slightly dependent of k. These longitudinal modes correspond to the plasma
with k. The correction term is very oscillations that we discussed qualitatively at the start of the section.
small, and arises from the breakdown
Figure 7.12(a) shows a schematic diagram of the longitudinal electron
of some of the approximations used in
this derivation. See Exercise 7.18. displacements within a plasma oscillation and the fields that they gen-
erate.
The existence of longitudinal solutions at the plasma frequency is a
consequence of the fact that r = 0 at ωp . In a medium with zero average
 7.5 Plasmons 201

Fig. 7.12 (a) Charge fluctuations in

a free carrier plasma oscillation. The
lighter regions denote areas with excess
electron densities. The small arrows in-
(a) (b) hwp dicate the direction of the electric fields,
which are parallel to the direction of
propagation of the wave, as indicated
by its wave vector k. (b) Excitation of
Ein Eout plasmons by inelastic scattering of par-
hwp ticles. The case in which two plasmons
are excited is shown. For metals, elec-
trons with keV energies are used, but
k sample for doped semiconductors, optical fre-
quency photons have sufficient energy.

charge density, Gauss’s law (eqn A.10) combined with eqn A.3 tells us
that
∇ · D = ∇·(r 0 E) = 0 . (7.47)
If r = 0, we then deduce that ∇ · E = 0. This is the normal situa-
tion for transverse electromagnetic waves in which the electric field is
perpendicular to the direction of the wave. However, if r = 0, we can
We shall come across another exam-
satisfy eqn 7.47 with waves that have ∇ · E = 0, i.e. longitudinal waves. ple of longitudinal modes at frequencies
We thus conclude that a dielectric can support longitudinal electric field where r = 0 when we consider phonons
waves at frequencies that satisfy r (ω) = 0. in Chapter 10. (See Section 10.2.2.)
Equation 7.44 shows us that the longitudinal oscillations of the plasma
behave as harmonic oscillators with a natural resonant frequency at ωp .
The derivation is completely classical, and the oscillator can have any
energy. However, we know in fact that the energy of harmonic oscillators
is quantized. We therefore expect the energy of the plasma oscillations
to be quantized in units of
ωp . The quasi-particles that correspond to
these quantized plasma oscillations are called plasmons. As shown in
eqn 7.46, the frequency of the plasmons is independent of their wave
vector.
Since plasmons are associated with longitudinal plasma oscillations,
they cannot be excited directly by light, which is a transverse wave. In-
stead, they have to be observed by techniques of inelastic scattering, in
which a beam of particles excites plasmons while passing through the
medium, as illustrated in Fig. 7.12(b). The energy Ein of the incoming
particles must be significantly larger than the plasmon energy. Conser-
vation of energy requires that:

Eout = Ein − n
ωp , (7.48)

where Eout is the energy of the outgoing particle, and n is the number
of plasmons emitted. The detection of particles with energies given by
eqn 7.48 establishes that plasmons have been excited.
In the case of metals, the plasmon energies are several eV, and so
electrons with keV energies are typically used. By measuring the en-
ergy spectrum of the electrons emerging from a thin sample, the plasma
202 Free electrons

Raman scattering rate

plasmon plasmon
emission absorption

Fig. 7.13 Raman scattering measure-

ments on n-type GaAs at 300 K. The
doping density was 1.75 × 1023 m−3 .
The data are displayed as a function
of the energy shift of the outgoing
photons relative to the incoming ones
in wave number units. After Moora-
-400 -200 0 200 400
dian (1972),
c Excerpta Medica Inc., Energy shift (cm -1)
reprinted with permission.

frequency can be determined. This technique is called electron energy-

loss spectroscopy.
Plasmons can also be observed in doped semiconductors. Since the
plasma frequencies are much lower, it is possible to use inelastic light
scattering techniques (i.e. Raman scattering) to measure the plasmon
energies. The general principles of Raman scattering will be discussed
in Section 10.5. The basic point is that the energy
ωout of the outgoing
photon must satisfy:

ωout =
ωin ±
ωp , (7.49)
where
ωin is the energy of the incoming photon. The + sign corresponds
to plasmon absorption and the − sign to plasmon emission. Plasmon
absorption is possible here, but not in the case of metals, because the
plasmon energies are comparable to the thermal energy kB T . This means
that there might already be plasmons excited in the sample before the
incident photon arrives, and thus there is some probability that a plas-
mon might be destroyed and its energy transferred to the photon.
Two weak peaks at ±272 cm−1 and Figure 7.13 shows the results of a Raman scattering experiment on
±296 cm−1 are also present in Fig. 7.13. n-type GaAs at 300 K. The doping density was 1.75 × 1023 m−3 . The
These are caused by optical phonons.
(See Section 10.5.2.) The phonon sig-
Raman intensity is plotted as a function of the frequency shift of the
nals are linearly polarized, and have light in wave number units. The data show two clear peaks shifted by
been strongly suppressed in the data by ±130 cm−1 relative to the incoming laser beam due to plasmon emission
the use of orthogonal polarizers in front and absorption. The electron effective mass of GaAs is 0.067m0 and opt
of the detector. Note that it is very
common to use wave number units in is 10.6. Hence from eqn 7.24 we find ωp = 2.8 × 1013 rad/s, which is
Raman spectroscopy. The wave number equivalent to 150 cm−1 . The experimental data are thus in reasonably
ν is equal to the reciprocal of the wave- good agreement with the model.
length: ν = 1/λ. It is effectively a unit
of energy with 1 cm−1 ≡ 0.124 meV.
7.5.2 Surface plasmons
Careful analysis of the electron energy-loss spectra from a metal typically
reveals that there are two different types of plasmon within the metal,
namely bulk and surface plasmons. We have considered the first type
in the previous subsection, and our task now is to explain the second.
Interest in these surface plasmons has increased dramatically in recent
years, and a new field of research has burgeoned called plasmonics.
 7.5 Plasmons 203

Surface plasmons are quantized electromagnetic surface waves that

are localized at the interface between a plasma and a dielectric material. dielectric
We are interested here in the case where the plasma is a metal. The
dielectric is usually the air, although it might also typically be glass or a
semiconductor. The waves propagate along the interface plane, and the
electron charge density fluctuations in the metal generate electric field
lines as shown in Fig. 7.14. The amplitude of the electric field decays
metal
exponentially on either side of the interface.
It is apparent from Fig. 7.14 that the surface plasmons have both
transverse and longitudinal electric field components, which contrasts Fig. 7.14 Electric fields associated
with bulk plasmons which are purely longitudinal. The presence of the with electron charge density fluctua-
transverse component means that surface plasmons can interact directly tions at the surface of a metal.
with photons. This interaction is sufficiently strong that we need to con-
sider the photon and plasmon as a coupled system called a polariton. In
general, polaritons are coupled electric polari zation–photon waves. Sev-
eral different types of polariton are possible, and the type of polariton
that we are considering here is called a surface plasmon polariton. Phonon polaritons are considered in
The dispersion of the surface plasmon polaritons can be found by Section 10.3.
solving Maxwell’s equations. We define axes so that the plane z = 0
corresponds to the interface, with positive and negative z corresponding
to the dielectric and metal respectively, as shown in Fig. 7.15(a). The
wave is assumed to be propagating in the x direction. The electric field
has components in both the x and z directions, and its amplitude decays
exponentially as a function of the distance from the interface, as shown
in Fig. 7.15(b). In this geometry, the electric and magnetic fields can be
written in the form:
E d (x, z, t) = [E dx , 0, E dz ] ei(kx x−ωt) e−kz z ,
d d
In principle, it might also be possible to
have waves with electric and magnetic
d
−kzd z
H d (x, z, t) = [0, Hyd , 0] e i(kx x−ωt)
e , field components along the y and x di-
m
(7.50) rections respectively. However, it can be
x−ωt) +kzm z
E (x, z, t) = x , 0, E z ] e
m
[E m m i(kx
e , shown that these solutions are not pos-
m
x−ωt) +kzm z sible. See Maier (2007).
H m (x, z, t) = m
[0, Hy , 0] e i(kx
e ,
where the labels ‘d’ and ‘m’ refer to the dielectric and metal respectively.
Note that the opposite sign of the decay term in the z direction in
the dielectric and the metal ensures that the fields are localized at the
interface.
The fields given in eqn 7.50 must satisfy Maxwell’s equations and the
boundary conditions that apply when there is no net free charge density.
Consider first the boundary conditions. The tangential components of E
and H, together with the normal component of the electric displacement
D, must be continuous at the interface. On recalling that D = r 0 E,
we then have that:
E dx = E m
x ,
Note that the most interesting case
to consider is when ω < ωp , where
Hyd = Hym , (7.51) m is negative. E d
z and E z therefore
m

point in opposite directions, as shown

d E dz = m E m
z , in Fig. 7.14.
where d and m are the relative permittivities of the dielectric and metal
respectively. The requirement that these conditions apply along all the
204 Free electrons

z surface implies that

(a)
kxd = kxm ≡ kx , (7.52)
dielectric: ed
where kx is the common x component of the wave vector on both sides
z=0 x of the interface.
k Now consider Maxwell’s fourth equation (eqn A.13) with j = 0 and
metal: em D = r 0 E. The fact that E y = 0 and Hx = 0 allows us to relate Hy to
E x:
∂Hy ∂E x
z − = r  0 . (7.53)
∂z ∂t
(b) On substituting the fields from eqn 7.50, this gives:
½Ezd (z)½ kzd Hyd = −id 0 ω E dx ,
(7.54)
−kzm Hym = −im 0 ω E m
x .
0 ½Ez½
By using E dx = E m d m
x and Hy = Hy from eqn 7.51, we can rearrange this
½Ezm(z)½
to find:
kzd km
+ z = 0. (7.55)
d m
Fig. 7.15 (a) Definition of axes for the Note that this is consistent with both kzd and kzm being positive when m
interface between a dielectric medium is negative, i.e. when ω < ωp . Since the overall charge density is zero,
and a metal with relative permittivi-
ties of d and m respectively. The plane
the fields must satisfy (cf. eqns A.25 and A.28 with µr = 1):
z = 0 defines the interface, and the po-
lariton propagates in the x direction, as r ∂ 2 E
∇2 E = . (7.56)
shown by the k vector. (b) Exponential c2 ∂t2
decay of the field amplitudes as a func-
tion of the distance from the interface. On inserting the fields from eqn 7.50 and using eqn 7.52, this gives:
d 2
kx2 − (kzd )2 = ω ,
c2 (7.57)
m
kx2 − (kzm )2 = 2 ω 2 .
c
Then, on using eqn 7.55 to eliminate the kz components, we finally
obtain:

1/2 √
1/2
ω m d ω d m
kx = = . (7.58)
c m + d c m + d
This equation gives the dispersion curve for the surface plasmon polari-
tons.
Figure 7.16 compares the dispersion of the surface plasmon polaritons
to the photon dispersion in the bulk of a metal with a dielectric constant
given by eqn 7.7, namely:

ωp2
m = 1 − . (7.59)
ω2
The relative permittivity of the dielectric is assumed to be real and
independent of frequency.
Consider first the dispersion in the bulk of the metal, which is shown in
Fig. 7.16(a). For frequencies below ωp , the light is reflected, and there are
 7.5 Plasmons 205

w w
(a) (b)

w = ck w = ckx /Öòd
wp wp Fig. 7.16 (a) Photon dispersion in the
wsp bulk of a metal with a dielectric func-
tion given by eqn 7.7. (b) Surface plas-
mon polariton dispersion for the same
metal. The curve is drawn for the case
0 0 where the dielectric is air (i.e. d = 1).
0 1 2 0 1 2
ck /wp ckx /wp The dashed line in (b) shows the dis-
persion for light in the dielectric.

just evanescent fields in the medium. There are therefore no propagating

modes with ω < ωp . For ω > ωp , the dispersion is given by (see eqn 7.45):
 1/2
ω = ωp2 + c2 k 2 . (7.60)

This asymptotes to ω = ck at large wave vectors.

The situation for the surface plasmon polaritons is qualitatively differ-
ent, as shown in Fig. 7.16(b). The ‘light line’ for the dielectric defined by
√
ω = ckx / d is shown for comparison. Three distinct frequency regions
can be identified.
√
(1) 0 < ω < ωp / 1 + d . In this frequency region, both m and (m +
d ) are negative, so that kx is real. For small ω, |m | is large.
Therefore the plasmon dispersion curve approaches the light line
for small kx .
√
(2) ωp / 1 + d < ω < ωp . In this region m is negative, but (m + d )
is positive. kx is therefore imaginary, and there are no propagating
modes.
(3) ω > ωp . Both m and (m + d ) are now positive, so that real
√ m →√1, and
solutions for kx are again found. At high frequencies,
the dispersion approaches the limit with ω = ckx 1 + d / d .

In region (1) at large kx the group velocity (i.e. dω/dk) is zero and
ω → ωsp . This asymptotic frequency limit is called the surface plasmon
frequency. Equation 7.58 shows us that kx → ∞ when (m +d ) → 0, and
so we can find ωsp by solving m (ω) = −d . For an undamped plasma
with m (ω) given by eqn 7.59, we find: In electron energy-loss experiments on
metals, it is common to observe peaks
ωp corresponding to both bulk and sur-
ωsp = √ . (7.61) face plasmons. This gives a convenient
1 + d
method for measuring both ωp and ωsp .
√
Note that ωsp = ωp / 2 when the dielectric is air.
The behaviour of the surface plasmon polaritons in region (1) where
ω < ωsp is the most interesting, since this corresponds to propagating
modes for frequencies below the plasma frequency. The spatial extent of
the fields in the z direction can be found from the values of kz in the
206 Free electrons

metal and the dielectric. This can be done by substituting eqn 7.58 back
into 7.57 to obtain:

1/2
d ω −2d
kz = ,
c  m + d

1/2 (7.62)
m ω −2m
kz = .
c  m + d
Since (m + d ) goes to zero at ωsp , the decay constants increase with
ω and diverge on approaching ωsp . The field decay length lz is equal
to 1/|kz |, and this implies that the polaritons become more localized
as ω approaches ωsp . Note that it is usually the case that |kzm | > |kzd |,
and hence that the plasmon extends further in the dielectric than in the
metal, as illustrated in Fig. 7.15. At optical frequencies, lzd and lzm are
typically a few hundred or few tens of nanometres, respectively. (See
Exercise 7.20.)
Two other branches of plasmonics deal One of the aims of the research field of plasmonics is to propagate
with the enhancement of radiative ef- electromagnetic waves in a metal as surface plasmon polaritons. With
ficiencies and the increased transmis- values of lzm being in the sub-100 nm range, the waves can be confined
sion of light through sub-wavelength
apertures. In the first case, the radia-
to dimensions smaller than the wavelength of light, and we then enter
tive efficiency of an atom in a dielec- the realm of nanophotonics, which is not accessible to conventional
tric can be enhanced by placing it close optics due to diffraction limits. The distance that the plasmon modes
to a metal surface, and hence exploit- can propagate along the surface is determined by the imaginary part of
ing the large electric field amplitude
that is present at the interface. In the m . This can be several tens of micrometres (see Exercise 7.20), which
second case, it has been demonstrated is more than adequate for the applications that are being considered.
that plasmonic effects can enhance the An issue that has to be addressed in plasmonics is the way to couple
transmission of light through periodic light to the polaritons. It is apparent from Fig. 7.16(b) that the polariton
arrays of sub-wavelength-size holes in
an optically thick metallic film. Details modes always lie below the light line. This means that polaritons can
of these and other applications may be never transform directly into light: they are non-radiative modes. By the
found in the Further Reading. same token, it is not possible to couple light directly from the dielectric
into the polariton modes, since it is never possible to match the wave
vectors. Therefore, in order to couple external light into the surface
plasmon polaritons, techniques must be used to change the wave vector
of the light. One way to do this is to use a grating. In fact, it has been
known for a long time that the reflectivity of metallic ruled gratings can
drop significantly when one of the diffracted orders propagates parallel
to the surface. This effect, which is called Wood’s anomaly, is now
known to be caused by the excitation of surface plasmon polaritons in
the metal. Details of how the coupling to polariton modes is achieved in
practice may be found in the references cited for Further Reading.
A striking example of surface plasmon effects is to be found in consid-
Colloids can be solids, liquids, or gases.
The particle size in a colloid should
ering the optical properties of metal colloids. Metal colloids are made by
be smaller than the wavelength of dispersing a large number of very small metallic particles (i.e. ‘nanopar-
light. An interesting application of ticles’) throughout a homogeneous medium such as glass or water. It
metal colloids is in the production of has been known from antiquity that the colour of metal colloids is dif-
stained glass. The colouration of me-
dieval stained glass is typically caused
ferent from that of the bulk metal, and this phenomenon is now known
by gold, silver or copper nanoparticles to be caused by the resonant excitation of surface plasmons in the metal
incorporated into the glass during the nanoparticles. However, in contrast to surface plasmon polaritons, the
melt process. surface fields in the metallic nanoparticles do not propagate, since they
 7.6 Negative refraction 207

1.2

Absorption (arb. units)

0.8
Fig. 7.17 Absorption spectrum for a
thin film of gold nanoparticles embed-
0.4 ded within an organic dielectric with
d ≈ 2.5. The film was grown on a glass
substrate, and the spectrum was mea-
sured at room temperature. The parti-
0.0 cle size was 6–7 nm. Unpublished data
300 500 700 900 from M.R. Sugden and T.R. Richard-
Wavelength (nm) son.

are confined to the nanoparticles. They are therefore known as localized

surface plasmons.
A simple explanation for the change of the optical properties of col-
loidal metals compared to the bulk can be given in terms of the polar-
izability. If the particle radius a is much smaller than the wavelength of
the light, then it can shown that the polarizability is given by: The derivation of eqn 7.63, which has a
similar form to the Clausius–Mossotti
m − d relationship given in eqn 2.35, may be
α = 4πa3 , (7.63)
m + 2d found, for example, in Maier (2007).
The general treatment of the inter-
which has a resonance when√ m = −2d . In an undamped plasma, the action of light with conducting metal
resonance occurs at ωp / 3 when the dielectric is air (see Exercise 7.21), spheres is called Mie theory: see Born
and is independent of the particle size. However, the resonance in real and Wolf (1999).
metals is shifted by interband absorption, and does depend somewhat
on the size. Figure 7.17 shows the absorption spectrum of a thin film of
gold nanoparticles embedded within an organic dielectric with d ≈ 2.5.
A strong plasmonic absorption peak centred at 580 nm (2.1 eV) is clearly
resolved in the data. When the nanoparticles are suspended in water, the
resonance occurs at higher frequencies owing to the lower permittivity
of the dielectric. (See Exercise 7.22.) The absorption resonance is in fact
in the green spectral region, and the colloid appears red, instead of the
usual golden colour of the bulk metal. Similar effects can be observed
for other metals.

7.6 Negative refraction

Throughout this chapter we have been studying the optical properties
of materials that have negative values of r . We now wish to consider
briefly the properties of materials that also have negative values of the
relative magnetic permeability, µr . As we shall see, this possibility leads
to the striking concept of a negative refractive index, which is a subject
that has attracted much interest in recent years.
The general relationship between the refractive index of a medium
and µr follows directly from Maxwell’s equations. Equation A.29 shows
√
that the speed of light in a medium is equal to c/ r µr , and so we can
write:
√
ñ = r µr . (7.64)
208 Free electrons

In everything that we have been considering so far in this book, we have

√
been assuming that µr = 1, so that eqn 7.64 reduces to ñ = r . In
fact, (µr − 1) characterizes the magnetic response of the medium, and
magnetic dipoles respond to oscillating electromagnetic fields in much
the same way that electric dipoles do. However, the natural resonant
frequencies are low (∼GHz at most), and so the magnetic response is
usually negligible at optical frequencies. It is therefore important to clar-
ify at the start that there are no known natural materials that have µr
significantly different from unity at optical frequencies. The materials
that we shall be considering here are therefore purely artificial.
The four possible general combinations of values of r and µr are
depicted schematically in Fig. 7.18. In quadrant I, both r and µr are
positive. This is the usual situation in a transparent optical medium, and
mr the solutions to Maxwell’s equations are travelling waves with a phase
velocity determined by a refractive index n that is both real and positive.
In quadrant II, µr is positive, but r is negative. This is the scenario
II I that we have considered at length in this chapter, since it applies to
ñ = ik n>0 the case of a metal below its plasma frequency. In these conditions, ñ
k>0 k=0 is purely imaginary. This means that the waves decay exponentially in
òr the medium: there are no propagating solutions, and incoming waves
n<0 ñ = ik from the air are reflected. A similar situation would occur in quadrant
k=0 k>0 IV, where r is positive, but µr is negative. The final case to consider
III IV is that which corresponds to quadrant III in which both r and µr are
negative. The properties of materials that fall into quadrant III were
first considered theoretically by Veselago in 1968.
Veselago’s main conclusion was that a medium with both r and µr
Fig. 7.18 Real and imaginary parts of negative would be transparent, and have a negative refractive index. Not
the complex refractive index for four surprisingly, this gives rise to many unusual properties. The k vector of
different combinations of values of r the wave and the direction of energy flow are in opposite directions.
and µr . The negative refractive index Since the energy flow is determined by the Poynting vector E × H, the
regime occurs in quadrant III, where
both r and µr are negative. reversal of the direction of energy flow relative to k is consistent with
reversing the direction of the magnetic field. This means that E, H
and k now form a left-handed system instead of the usual right-handed
arrangement, and so materials with n < 0 are sometimes called ‘left
handed’.
Even more striking effects occur when light is refracted on entering
the negative index medium. If the incoming ray has an angle of incidence
θi , then the angle of refraction θr is given by Snell’s law:
sin θi
= n. (7.65)
sin θr
When n is negative, θi and θr have opposite signs, as shown in Fig. 7.19(a).
This leads to the possibility that the medium can behave as a lens, as
shown in Fig. 7.19(b). An object to the left of the medium is brought
to a focus to the right, after passing through an intermediate focus. The
trajectory of the rays only depends on the thickness of the medium, and
there are therefore no aberrations. Hence the medium is said to behave
as a perfect lens. Since the rays emerge exactly as they would from the
 Chapter summary 209

source, the medium is invisible. n=1 n<0

Interesting though these properties might be, there would be no point (a)
considering them if it were not possible to obtain a medium that has qi qr
n < 0. Metals such as silver or gold have negative values of r at most
frequencies, and so the issue becomes that of obtaining the negative
value of µr . As mentioned above, there are no known natural materi-
als that possess this property at useful frequencies. Hence the negative
refractive index must be engineered by creating an artificial structure
called a metamaterial. It was not until the mid-1990s that this subject
became of widespread interest, following work by Pendry in which the
first practical designs for metamaterials with a negative refractive index (b)
were proposed.
The principle behind a metamaterial is to create a metallic structure n ® -1
designed to behave like a magnetic dipole resonator. Figure 7.19(c) illus-
trates one of the standard designs considered in the literature, namely
an array of split rings. In this case the magnetic response is determined
by the design of the constituent units (i.e. the split rings), which must
be smaller than the wavelength of the electromagnetic waves, but are
still much larger than the underlying atoms. It is for this reason that
the medium is called a metamaterial. (c)
We have seen in Chapter 2 that the refractive response of an electric-
dipole resonator is negative above the natural frequency ω0 . (See, for
example, Fig. 2.6.) Magnetic resonators behave in a similar way, and
so we can expect to obtain µr < 0 in the frequency region above their
resonant frequency. The difficulty is that the value of ω0 depends on
the size of the structure. A split-ring array with a period of ∼ 10 mm
has negative refraction at GHz frequencies, but much smaller structures Fig. 7.19 (a) Negative refraction in a
must be used for optical frequency experiments. For this reason, the medium with n < 0. (b) Perfect lens-
ing for n < 0. (c) Split ring design for
underlying principles of negative index materials are usually tested first
producing a negative refractive index.
at microwave frequencies. Further details of the design of metamaterials In (a) and (b) the arrows give the di-
and progress in obtaining negative refraction at optical frequencies may rection of the Poynting vector.
be found in the works cited for Further Reading.

Chapter summary

• Free electron effects are observed in metals and doped semicon-

ductors. They can be modelled by the classical dipole oscillator
model with no restoring force term. This approach is called the
Drude–Lorentz model.
• The free electron plasma reflects strongly up to the plasma fre-
quency, which depends on the electron density. The damping rate
of the oscillations is determined by the momentum scattering time
deduced from electrical conductivity measurements.
210 Further reading

• Metals reflect strongly due to the plasma reflectivity effect. At

frequencies above the plasma frequency, the metals become trans-
parent. This effect is called the ultraviolet transparency of metals.
• Interband transitions are possible in metals from states below the
Fermi energy to empty levels above it. The interband absorption
can reduce the reflectivity from the value predicted by the Drude–
Lorentz model, and must therefore be considered to obtain a good
fit to experimental reflectivity data.
• Doped semiconductors reflect at frequencies in the infrared due to
the free electrons and holes generated by the doping process. Free
carrier absorption can be observed at frequencies above the plasma
frequency but below the fundamental absorption edge at the band
gap.
• P-type semiconductors show an additional absorption mechanism
in the infrared due to intervalence band transitions.
• Doped semiconductors show sharp infrared absorption lines due to
impurity transitions at low temperatures. At room temperature,
the impurity states broaden the fundamental absorption edge.
• Plasma oscillations can occur at the plasma frequency. The quan-
tized oscillations are called plasmons. These can be observed by
electron energy-loss spectroscopy in metals, or by Raman scatter-
ing in doped semiconductors.
• Surface plasmons correspond to localized electromagnetic fields at
the interface between a metal and a dielectric. Propagating surface
plasmon polariton modes can be excited by coupling light to the
metal with a grating or prism.
• Materials in which both the relative permittivity and magnetic
permeability are negative are characterized by a negative refractive
index. The artificial structures that show these effects are called
metamaterials.

Further reading
The properties of electromagnetic waves in a conducting sorption.
medium are covered in many electromagnetism and op- The properties of plasmons are treated in depth in
tics textbooks, for example Bleaney and Bleaney (1976), Maier (2007). Bulk plasmons are covered in Kittel (2005),
Born and Wolf (1999), or Hecht (2001). while the classic text on surface plasmons is Raether
The free carrier model of metals is covered in Singleton (1988). Review articles on the research fields of plas-
(2001). It is also covered in Ashcroft and Mermin (1976), monics and nanophotonics may be found in Barnes et
Burns (1985), or Kittel (2005). al. (2003), Lal et al. (2007), Maier and Atwater (2005),
Free carrier reflectivity and absorption in semiconduc- Murray and Barnes (2007), or Ebbeson et al. (2008).
tors has been reviewed by Pidgeon (1980), and is also The concept of negative refraction was proposed in
covered by Yu and Cardona (1996). Yu and Cardona give Veselago (1968). Introductory reviews on the subject may
further details about intervalence band and impurity ab- be found in Pendry (2004) or Pendry & Smith (2004).
 Exercises 211

A more detailed review is given in Ramakrishna (2005). quencies is reviewed in Shalaev (2007).
Progress on obtaining negative refraction at optical fre-

Exercises
(7.1) Derive a relationship between the Fermi energy EF (7.6) Estimate the fraction of light with wavelength
of a metal and its plasma frequency ωp . 1 m that is transmitted through a 20 nm thick
(7.2) The ionosphere reflects radio waves with frequen- gold film at 77 K, where the DC electrical conduc-
cies up to about 3 MHz, but transmits waves with tivity is 2 × 108 Ω−1 m−1 . The plasma frequency
higher frequencies. Estimate the free electron den- and electron density of gold are given in Table 7.1.
sity in the ionosphere. (7.7) Figure 7.20 shows the measured reflectivity of gold
(7.3) Estimate the skin depth of radio waves of fre- in the wavelength range 100–1000 nm. Account
quency 200 kHz in sea water, which has an av- qualitatively for the shape of the spectrum, and
erage electrical conductivity of about 4 Ω−1 m−1 . deduce the energy gap between the d bands and
Hence discuss the difficulties you might encounter the Fermi energy. Use the data to explain the char-
when attempting to communicate with a sub- acteristic colour of gold.
merged submarine using radio waves. (7.8) What is the value of the dielectric constant of a
(7.4) Cesium metal is found to be transparent to electro- medium that has zero reflectivity? Use eqn 7.22
magnetic radiation of wavelengths below 440 nm. to show that the reflectivity of a lightly damped
Calculate a value for the electron effective mass doped semiconductor is zero at the angular fre-
using the data given in Table 7.1. quency given in eqn 7.25.

(7.5) The momentum scattering time of silver is 4.0 × (7.9) Use the data shown in Fig. 7.7 to deduce the value
10−14 s at room temperature. Calculate the dielec- of the electron effective mass of InSb at each car-
tric constant at 500 nm, neglecting interband ab- rier density. Take
opt = 15.6.
sorption effects. Hence estimate the reflectivity of (7.10) The absorption coefficient at room temperature of
a silver mirror at this wavelength. See Table 7.1 an n-type sample of InAs with a doping level of
for the plasma frequency of silver. 1.4×1023 m−3 is found to be 500 m−1 at 10 m. Es-
timate the momentum scattering time, given that
the electron effective mass is 0.023 m0 and the re-
1.0 fractive index is 3.5.
gold (7.11)∗ A laser beam operating at 632.8 nm with an inten-
0.8 sity of 106 W m−2 is incident on a sample of pure
Reflectivity

InP at room temperature. The absorption coeffi-

0.6 cient at this wavelength is 6 × 106 m−1 , and the
carrier lifetime is 1 ns. Estimate the free carrier
0.4 absorption coefficient at the wavelength of a CO2
laser (10.6 m), where the refractive index is 3.1.
0.2
The effective mass and momentum scattering time
for the electrons are 0.08 m0 and 2 × 10−13 s, while
0.0
200 400 600 800 1000 the equivalent values for the holes are 0.6 m0 and
5 × 10−14 s.
Wavelength (nm)
(7.12)∗ Consider the intervalence band processes illus-
Fig. 7.20 Reflectivity of gold in the wavelength range trated in Fig. 7.9 for a heavily doped p-type sam-
100–1000 nm. Data from Lide (1996). ple of GaAs containing 1 × 1025 m−3 acceptors.

∗ Exercises marked with an asterisk are more difficult.

212 Exercises

The valence band parameters for GaAs are given (7.18) In a metal, the frequency at which
r = 0 varies
in Table D.2. slightly with the wave vector k ≡ |k|, and for small
(a) Work out the Fermi energy in the valence band k we have:
on the assumption that the holes are degenerate.  3vF2 k 2

What are the wave vectors of the heavy and light ω(k) ≈ ωp 1 + ,
10ωp2
holes at the Fermi energy?
(b) Calculate the upper and lower limits of the where vF is the Fermi velocity of the electrons.
photon energies for the three absorption processes Consider a metal with electron density N =
labelled (1), (2), and (3) in Fig. 7.9, namely the 1029 m−3 and lattice constant a = 4 Å. Calculate
lh → hh, the SO → lh and the SO → hh transi- the relative size of the term in k2 for k = 0.1 π/a,
tions. i.e. 10% of the size of the Brillouin zone.
(7.13) Figure 7.11 shows the infrared absorption spec- (7.19) In an electron energy-loss experiment on a metal,
trum of n-type silicon, which has a dielectric con- two series of peaks are observed that obey eqn 7.48
stant of 16. Two series of lines labelled np0 and with ~ωp = 10.3 eV and 15.3 eV. Use the data in
np± are identified in the data. Table 7.1 to determine which metal is being inves-
(a) Show that the np0 series is consistent with tigated, and account for the two series of peaks.
eqn 7.30, and deduce a value for the electron ef- (7.20) A surface plasmon mode with a frequency corre-
fective mass for these transitions. sponding to a vacuum wavelength of 600 nm is ex-
(b) Show that the np± series follows the following cited at the interface between silver and air. The
formula: relative permittivity of silver at this frequency is
R∗ R∗
hν = 20 − ± , given approximately by
m = −18 + i.
1 n2 (a) Calculate the decay constants kzd and kzm , and
∗ ∗
stating the values of R0 and R± deduced from the hence deduce the field decay lengths in the direc-
data. tion normal to the surface in both the air and in
(7.14) It is found that the infrared absorption spectrum the metal.
of a lightly doped n-type semiconductor with a (b) The propagation length in the direction par-
relative permittivity of 15.2 consists of a series of allel to the surface is defined as the distance over
sharp lines at low temperatures. The energies of which the intensity drops by a factor of e−1 . Cal-
the lines are given by: culate the imaginary part of kx , and hence deduce
the propagation length for silver at 600 nm.
E(n) = R∗ (1 − 1/n2 )
(7.21) Show that the resonance of the polarizabil-
where R∗ is 2.1 meV and n is an integer greater ity of colloidal metal nanoparticles that obey
than 1. Explain why the energies of the absorp- eqn √
7.7 occurs at an angular frequency given by
tion lines are almost independent of the type of ωp / 1 + 2n2 in a medium with a refractive index
impurity atoms used for the doping, and deduce a of n. Evaluate this frequency for the case where
value for the electron effective mass. the dielectric is air.
(7.15) The fundamental absorption edge of a semicon- Table 7.3 Complex relative permittivity of gold be-
ductor shifts from 5.26 m to 5.44 m when doped tween 500 and 550 nm. Adapted from Raether (1988).
with acceptors. Deduce a value for the ground
state acceptor level energy relative to the valence Wavelength (nm) 1 2
band.
500 –2.3 3.6
(7.16) The beam from an argon ion laser operating at 510 –3.0 3.1
514.5 nm is incident on an n-type GaAs sample. 520 –3.7 2.7
A peak is observed in the intensity of the scat- 530 –4.4 2.4
tered light at 534.3 nm. Explain this observation, 540 –5.2 2.2
550 –6.0 2.0
and estimate the electron density, given that m∗e =
0.067 m0 and n = 3.3.
(7.17) Calculate the doping density at which the plas- (7.22) (a) The complex relative permittivity of gold in the
mons in n-type GaAs have the same wave number range 500–550 nm is given in Table 7.3. Use this
as the longitudinal optic phonon at 297 cm−1 . Take information to estimate the resonance wavelength
m∗e = 0.067 m0 and n = 3.3. of colloidal gold nanoparticles in water, which has
 Exercises 213

a refractive index of 1.33. diction of the previous exercise. Account for any
(b) Use the data for gold given in Table 7.1 to com- difference.
pare the result obtained in part (a) with the pre-